Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations Paper โข 2510.23607 โข Published Oct 27 โข 172